Machine Learning in ImageJ/Fiji - StarDist

[@donTataf]

Hello Craig, thanks for the great video! I am new to the world of image processing. I will be starting a project to identify fungal spore cells from images taken of microscopy slides (fluorescent and/or polarized). The goal is to train a model which can identify spores between different fungal species. In my case, the spores of the two species of interest differ slightly by size and spore wall structure. Essentially, my end goal is to process thousands of images of spores with a working model (1. identify the spores on a slide image, and 2. determine the species based on known parameters unique to each species). My apologies for the long winded inquiry, but any pointers or recommendations you could provide me would be immensely helpful. I am trying to determine different softwares and/or programs where this could be possible.

[@GuihuaZheng-zd9dp]

Hello, thank you for your video.Your video deserves to be seen by too many people. Please, Are these models only for the nucleus, can they be used in the shape analysis of mineral particles? I tried it and it did not work

[@CraigDaly]

Hi, i think it should work for any structure that has objects of a similar size. I'd guess the algorithm doesn't know its a nucleus just a shape of a particular size.

[@fburton8]

Impressive! It'll be interesting to see how easy it is train StarDist to detect the long thin smooth muscle nuclei.

[@CraigDaly]

Check out ‘ZeroCostDL4mic’ it uses google Colab and Jupyter notebooks to create, what looks like, a fairly straight forward training environment. I’ll try that along with Deep ImageJ and the StarDist scripts to figure out which is the easiest approach. There are some great trained models in Deep ImageJ that are worth checking out. Might do that for episode 2.

[@carlessanchez5527]

Hi Craig, I am trying to count the CSA of the muscle NADH images. With HE I have no problem, but when I want to differenciate the CSA of dark and pale fibers form the muscle I can't run Stardist. Would you have any suggeriment to differentiate the area of dark and pale fibers? thanks

[@carlessanchez5527]

Star dist doesn't get my photo. Just it takes the small pictures. Coult be that my pc is powerful not enough? or there is some trick to evaluate larger images?

[@CraigDaly]

Hi, not sure of the limitations of stardist. I would suggest you begin by converting your images to 512x512 (size) and 8-bit (Type). That might get you going. C

[@zova4875]

Thanks for the video! I have images stained for TH+ axons in the cortex and I need to quantify axons. Most axons need interpolation to be detectable as an axon. Do you recommend any plugins for this aim in 2D space? Thank you

[@CraigDaly]

Hi, have a look at the Weka Segmentation plugin. I have a short tutorial on that which might help. You need to train the algorithm but it's quite straight forward. [kzbin.info/www/bejne/bajGZmOZqcysorM]